Accelerate Protein Engineering with StaB-ddG Now on Vecura

What is StaB-ddG?

StaB-ddG is an innovative deep-learning model designed to predict the impact of point mutations on protein-protein binding free energy (ΔΔG, kcal/mol) directly from 3D structures. By treating binding energy as the difference between the complex’s folding energy and the individual partners' folding energies, it leverages a fine-tuned inverse-folding model (ProteinMPNN) as a robust proxy. This approach allows it to achieve accuracy comparable to traditional physics-based tools like FoldX while delivering results over 1000 times faster.

It helps researchers rank mutations by their predicted effect on affinity without needing time-intensive wet-lab assays or complex simulations. It is especially useful for antibody engineering, interface redesign, and triaging large datasets to identify hotspot positions for stabilizing or destabilizing protein interactions.

What can users do with StaB-ddG on Vecura?

With StaB-ddG on Vecura, users can:

• Perform single-complex scoring for specific point mutations on wild-type PDB structures.

• Execute high-throughput batch mutational scanning across extensive libraries using PDB archives and CSV inputs.

• Customize inference precision by adjusting Monte Carlo sampling parameters to balance speed and variance.

• Seamlessly process various protein-protein interfaces, including antibody-antigen and enzyme-substrate systems.

What the output means

The output provides a comprehensive CSV table containing predicted binding ΔΔG values (kcal/mol) for each input mutation. These predictions allow for the immediate ranking of variants to prioritize candidates for experimental validation.

This output should be used to support scientific decision-making. It does not replace experimental validation. Users should be aware of a documented ambiguity regarding the sign convention (negative vs. positive ΔΔG) in the original codebase and are advised to validate results against a known benchmark before interpretation.

Why this matters

The ability to rapidly predict the energetic consequences of interface mutations is a significant bottleneck in protein engineering and therapeutic design. Traditional physics-based methods often require substantial computational resources and time, limiting the scope of in silico screening.

By delivering deep-learning performance that rivals established benchmarks at a fraction of the wall-clock time, StaB-ddG empowers researchers to explore much larger mutational landscapes. This acceleration facilitates faster iterations in antibody discovery and structural biology, enabling more efficient prioritization of protein variants for laboratory synthesis and testing.

• Developed by: Research group associated with the ICML 2025 publication.

• Source: Official GitHub Repository

• Reference: StaB-ddG paper (ICML 2025)